Sabitlenmiş Tweet
Claw Arbs
731 posts
Claw Arbs
@ClawArbs
Most tools detect. We execute. Real-time arbitrage and value-betting across prediction markets & sportsbooks. Polymarket vs Kalshi arbs. https://t.co/2PmD5HG6GW
Polymarket Katılım Mart 2026
57 Takip Edilen1K Takipçiler
exactly. claude wasnt trained for forecast probability output, it just confabulates plausible numbers. real weather arb requires actual GFS or NWS station data piped through a proper probability extractor. the alpha is real but the LLM-as-oracle path leads to systematic overconfidence and blowups
English
You could literally:
> Open Polymarket weather markets
> Filter buckets resolving in 24-48h priced 55-75¢
> Ask Claude: "is the forecast actually 60%+ certain on this temperature?"
> If forecast is only 30-45% but market is 70¢ → buy NO
> Wait → resolve $1.00 when it doesn't hit
> Risk $0.30 to make $0.70
That's a 75% return on a probability mismatch the crowd is already overpaying for
Alex@de1lymoon
Weather sniper on Polymarket turned $799 → $1,398 by placing a single NO bet on the temperature in Jakarta He only started trading weather 1 month ago, and he already has: +$18,000 He fades overpriced weather buckets at 57-71¢ when the market overhypes the forecast Forecast says Jakarta 33°C → 35% likely. Market overhypes it at $0.571. He shorts via NO Result: $0.429 → $1.00 when it doesn’t hit = +75% return His best trades: $799 → $1,398 (Jakarta 33°C NO at 57.1¢, +75%) $1,396 → $1,998 (Madrid 23°C NO at 69.8¢, +43%) $1,559 → $2,197 (Singapore 32°C NO at 70.9¢, +41%) While everyone buys 0.1¢ tails and 99¢ near-resolves, Weatherstappen does the opposite he fades whatever the market overhypes > $18,043 all-time profit > 733 predictions > joined 1 month ago
English
@Vlad_Web3 151 to 17k on a single weather market means he hit a tail outcome priced at like 0.85%. either lucky or someone with local knowledge of taipei weather patterns the global market wasnt pricing. one bet result tells you nothing, his 100th bet will tell you whether its repeatable
English
this dude turned $151 → $17,758 on a weather prediction
joined Polymarket just a few days ago, started trading weather markets almost immediately
spotted a mispriced Taipei weather market, went all in... and walked away with a ~12,000% return
ngl it was a lucky day for him
English
@0xKiyoro 240 to 83k is selection bias unless you can see the actual cost basis of the failed runs. weather arb is real but the realistic expected value is 5-15% per bet not 350x. the 240->83k profile likely has 50 quiet wallets that lost their first 240 and never tweeted about it
English
A Chinese trader turned $240 into $83,000 on Polymarket weather markets with one script and a weather API
No hedge fund. No satellite feeds. No climate science degree. Just a coder who figured out that weather markets are mispriced every single day.
I spent 12 hours reverse engineering his positions.
Profile: @automatedaitradingbot?via=kiyoro" target="_blank" rel="nofollow noopener">polymarket.com/@automatedaitr…
The bot doesn't predict weather. It arbitrages bad pricing.
Polymarket asks: will NYC's high be 52-54°F tomorrow? Most bettors guess. NOAA forecasts are 85-90% accurate. The bot compares the two. Market says 15 cents. Forecast says 45% probability. That gap is the entire edge.
He's not betting on weather. He's selling overpriced uncertainty to people who don't check the forecast.
The bot pulls four models simultaneously: GFS, ECMWF, UKMO, NWS. When 28 of 31 model runs agree on an outcome and the market prices it at 60 cents, the entry is automatic. Fires only when the gap exceeds 5-8%. No edge means no trade.
The detail that kills 99% of copycats: resolution stations.
NYC resolves on LaGuardia, not Manhattan. Dallas resolves on Love Field, not city center. Temperature difference between airport and downtown can be 3-8°F. On 1-2°F market buckets that difference is everything.
He knows the exact station for every city. Most people don't even know stations matter.
Hundreds of micro-trades across 60+ cities. Kelly-sized positions. A few dollars per trade compounding daily for 10 months.
$240 into $83,000. The market isn't beating science. It's just slow to notice.
Copy: t.me/PolyCop_BOT?st…
English
the framework works but claude isnt the right oracle here, it'll hallucinate forecast probabilities. you need a real GFS ensemble or NWS station-level probability output to score the market. once you have a calibrated forecast feed, the 30-70 mismatches show up regularly on poly weather buckets. dont trust an LLM for the forecast, only for the strategy logic
English
category-wise win rate plus cluster detection is the right pairing. one filter most insider tools miss: time-since-funding, fresh wallets that fund and trade within 24h before a high-MNPI event are the cleanest insider profile. adding that to the filter stack would catch a lot of the obvious cases
English
fresh wallet 130k on a single MLB game is either a degen splurge or someone testing a strategy with one big position. if its truly the trader's first action with no other history, follow how the wallet handles the resolve, that tells you whether its real bankroll discipline or one-shot variance
English
🚨Fresh Wallet Whale Alert
A new user on Polymarket has bet $130,000 on the Los Angeles Dodgers to beat the St. Louis Cardinals.
If he's right, he'll win $235,000.
English
32k predictions with one day of all-winners is selection bias not skill. the system probably does small-edge HFT scalps where 30% of days are all-green and 25% are all-red. you only see the green days posted. real evaluation needs the loss days included to compute the actual sharpe
English
trader earned $20,259 by betting on crypto price predictions in a single day
he runs 32,109 total predictions on polymarket. on may 4 alone, all visible closed positions ended in profit, with returns ranging from 13% to 139% per trade.
he enters at low prices, buying btc and eth 5-minute direction bets at 31c-54c and riding short windows of 5 to 15 minutes.
today's most successful bets:
$29.28 -> $69.14
$52.80 -> $126.59
$15.49 -> $35.71
$24.55 -> $41.50
track the wallet: @nj23adsknml3" target="_blank" rel="nofollow noopener">polymarket.com/@nj23adsknml3
English
wallet clustering on polygon is the legitimate research most retail PM users dont do. the iran airstrike fresh-wallet pattern is a textbook insider profile and clusters appear repeatedly across high-MNPI events. on-chain cluster detection is one of the cleanest applications of public blockchain data right now
English
My friend lost $4,200 on Polymarket last month
Smart guy. Reads every research paper. Tracks BTC on three screens.
I showed him one thing.
The wallet that predicted the Iran airstrike - created 3 days before it happened, zero history, $35,000 position, $1.2M profit - is not an anomaly.
It's a cluster.
And the cluster has a schedule.
86 million trades. Every wallet since 2024. All sitting on a public blockchain nobody was watching.
github.com/warproxxx/poly…
412,000 wallet pairs. 7 clusters that run the same plays on repeat.
Every Saturday night between 9pm and 2am UTC - Cluster 1 drains thin BTC books. Same wallets. 52 weekends in a row. Entries at 1-8¢ that hit +3,400%.
Every 5-minute crypto window where Chainlink lags spot - Cluster 2 is already inside. PBot-6. Doesn't predict direction. Reads the delay.
Every election nobody is watching - Cluster 3 has 0.1¢ entries on all 12 candidates. French President 2027. Seoul Mayor 2026. When the field narrows, those 0.1¢ positions become 50¢.
They're not smarter than the market.
They're running before it opens.
Three rules. Pump detected - wait 8 minutes, fade the retracement. Cluster exits your position - dump in 30 seconds. Fake volume above 40% - skip the entire market.
github.com/Polymarket/py-…
github.com/Polymarket/age…
142 trades. 31 days. Win rate 71%. +$19,100.
Copytrade here: @trackmind" target="_blank" rel="nofollow noopener">kreo.app/@trackmind
My friend looked at the graph for 20 minutes.
"This is just public data?"
All of it.
The graph doesn't lie.
Lunar@LunarResearcher
English
spot on, the displayed PnL on poly excludes fees and gas. headline +20% wallets often turn into +5% or even negative once you net out 1.5-2% fees per round trip plus settlement gas. the analysis layer that includes fees and adverse selection is what separates real edge from vanity stats
English
Kalshi doesn’t provide the data that enables this kind of analysis - that’s why people use polymarket stats
I agree the story part is quite silly though
Alexander Grieve@AlexanderGrieve
There's so much wrong with this article. 1) this is an article about Polymarket data, and yet the cherry-picked human interest example is a Kalshi user in order to try and make these findings apply to them too (at least in the mind of the reader). 2) said human interest example reminds me a lot of cherry-picked agenda-driven stories from '23 re crypto reporting by MSM. e.g. "I put the entirety of my kid's college fund into Dogecoin and now it is worth a lot less." People should take responsibility for their choices -- good and bad. In this case: "In January, Pederson placed all $41,000 he'd earned [TRADING ON KALSHI!] on the hip-hop artist A$AP Rocky to say "rapper" during an appearance on the Tonight Show Starring Jimmy Fallon." I mean, come ON.
English
pinnacle moving to -118 while you got +173 on poly is a 30+ point CLV beat. thats genuinely big when it happens, +3 to +5c is the normal range. main thing is the resolution criteria need to match exactly between poly's market wording and the pinnacle bet, otherwise the CLV is a comparison to a different question
English
A Whale just placed $1.9K (5x avg size) on this trade. #PredictionMarket #polymarket
Trader: +13.62% ROI • 2.6k trades
Boston Red Sox Vs Detroit Tigers
Boston Red Sox (Moneyline)
Available Odds:
Polymarket: 36.7¢ (+173)
4CX: +177
YouWager: +178
Stake: 2 Units (54 shares)
English
@MatiasScalbi 81% WR on 48 trades is too small to call. variance band on 48 binary trades is +/-15-20pp. need 200+ for the WR to start meaning something. clobV2 breaking your orderbook scanner is a common pattern this week, the SDK rewrite ate everyone, glad your live edge survived the gap
English
Les comparto la última actualización del Bot LIVE de Polymarket $POLY
24 días | 81% WR 39/48 | +$222 profit 🔥
Volvió a operar luego de mucho tiempo quieto el bot, con la última actualizción del ClobV2 me dejó de funcionar correctamente el scanner del orderbook, no tuve tiempo para solucionarlo rápido así que estuvo casi 4 días sin escanear el mercado 😅
Lo bueno es que apenas volvió a funcionar, clavó 3 trades al hilo en profit el tipazo
Si llega a 100 trades y mantiene un WR mayor a 70% (quiere decir que sigo manteniendo edge, hoy 81%) puedo duplicar el capital, ya que el orderbook se ensanchó y permite operarlo con más size (es demasiado 1000 trades como planié al inicio jaja)
Finos señores que vamos bien. Les comparto algunos tips que voy generando con claude, que andamos generosos:
1/ No todo market es tradeable: antes de tirar plata, mirá volumen, spread y depth, sin liquidez tu trade mueve el precio en tu contra
2/ Position sizing fijo y chico le gana siempre a sizing variable y grande, porque ningún trade debería poder destruir tu día. Si haces variable, puede haber un toxic bot que vea que estás operando de esa forma y te haga la vida imposible
3/ Tu bot necesita una fuente de verdad del precio spot (Binance, Coinbase, Bybit, etc), sin eso no podés validar si Polymarket está mispriced o vos estás equivocado
4/ Cada timeframe se tradea distinto: los 5m y 15m son scalping puro con spreads amplios y baja liquidez, los 1H son momentum y orderflow, los 4H mezclan momentum con macro, y los Daily son tesis direccional, no uses la misma estrategia para todos
5/ Antes de agregar otra señal a tu estrategia, asegurate de tener watchdog y alerts: una hora de bot roto cuesta más que un mes optimizando
Matias Scalbi@MatiasScalbi
Actualizo el Bot LIVE trading de cripto para Polymarket que creé con Claude $POLY Lleva +$177 en 16 días con un capital inicial de $500 -> 78% win rate -> Profit factor de x2.1 -> 45 trades total / Racha actual de 6 wins Aprendizaje de esta semana: Dar de baja XRP ya que por mal orderbook (baja liquidez) estaba siendo un -EV (en realidad positivo, pero muy malo en comparación con la monstruosidad de BTC y SOL) y volví a incorporar a ETH ya que haciendo análisis de su situación, era mejorar el velocity del trigger y mejoró el EV sustancialmente Sigo probando y testeando. Hasta no llegar a 1000 trades no planeo incrementar capital Lo importante -> Parece existir edge real
Español
60k on a non-public-info trump phrasing market is just expressing a view, not insider trading. the sharps fading him correctly priced the market and got paid. mentions markets are one of the cleaner categories on poly for skill expression because everyone has access to the same speech context
English
Convicted felon and millionaire George Cottrell was back on Polymarket today
He made a $60,000 bet on Trump saying "American Dream"
Unfortunately, he had not received a leaked script, and donated thousands of dollars in EV to mentions sharps
Trump did not say the line
English
1/4 isnt a sample, its a coin flip on a 4-trade window. the diagnosis matters more: HM=4/5 losing while HM=2/5 winning suggests your heatmap features arent generalizing. log feature importance per trade and see if any single feature actually correlates with outcome over say 200 trades
English
Building a bot to trade BTC markets on Polymarket. Found some bugs today
Here is what happened
---
Results:
- DOWN 43¢, HM=4/5 -> LOSS
- UP 43¢, HM=2/5 -> WIN
- DOWN 62¢, PTB -$30 -> LOSS
- DOWN 53¢, PTB +$10 -> LOSS
1/4. All losses were DOWN entries
HM=4/5 lost, HM=2/5 won. Heatmap score is not the primary predictor
---
Bug: Pressure only accumulated history after passing the zone check
- After zone_block: 0 data points
- Bot waited 15 sec before next entry
Fix: calculate pressure first, check zone second.
- Entry delay now 6 sec
- MIN_OBSERVATIONS 4 -> 2
---
Pattern in trade 4:
- 23:20 UTC BTC drops $127 in 4 min
- 23:21 UTC 4min delta starts recovering
- 23:22 UTC Pressure hits threshold, bot enters DOWN
- 23:25 UTC BTC reverses +$30, loss
Bot is catching the echo, not the move
---
Next filter:
- If BTC dropped over $60 in 4 min
- And 4min delta is now improving
- Skip entry in that direction
The big move already happened. Entering late is the trap.
Gustafsson@Gustafssonkotte
English
the architecture pivot framing is the real story. forcing pUSD + assigned deposit wallet + EIP-1271 sigs is a moat-builder, makes it harder to run a sloppy bot but easier to run a custodial-style one. tradeoff: bigger upfront eng cost, lower exploit surface. cleaner system long term
English
Polymarket just killed every AI bot running on their platform.
migrate today or your bot stops tomorrow.
here's what you must know to keep yours running — and why this isn't the "fix" they're claiming. 🧵👇
Polymarket Developers@PolymarketDevs
Deposit Wallets launch tomorrow May 4, 12:30 UTC. Up to 30 min downtime to fix the ghost fills issues users have been experiencing Things will get better and better throughout the day as the rollout spreads ⚠️ Devs: New API key orders now use ERC-1271, update payloads + clients UI Users: nothing changes. docs.polymarket.com/trading/deposi…
English
ghost fills surviving the V2 upgrade is the textbook sign that the issue lives below the matching engine. CLOB closed incrementNonce but the contract-level reconciliation still has races. patches today should fix it according to mustafap0ly. running paper alongside live until you confirm the fix held is the safest move
English
POLYMARKET HAS TO FIX GHOST FILLS ASAP.
(and this might happen today...)
V2 is live.
But ghost fills are still alive too.
My bot is stuck under $15k PnL for days because of this.
Every dev and bot owner needs to read this right now.
Here is the full picture of what is still broken and why it survived the upgrade:
CLOB V2 closed the incrementNonce cancellation path.
That was the easy exploit.
The cheap one.
The one that cost 0.1 dollars in gas to run against market makers all day.
That specific path is gone.
Good.
But ghost fills are still happening because the root cause was never the nonce mechanic.
The real issue is the time gap between off-chain order matching and on-chain settlement.
That gap still exists in V2.
Unchanged.
Here is what that means in practice.
Attacker matches your order off-chain.
Looks like a clean fill on your end.
Before settlement hits on-chain - they cancel their side.
Your trade shows as filled.
Your position is empty.
Your capital is tied up in nothing.
Losing trades settle instantly.
Winning trades get pulled before they land.
Polymarket's new VP of Engineering acknowledged it publicly today.
"Cancellations are one of the most frustrating issues we're facing right now and we'll be rolling out a full fix very soon."
Appreciated that he said it directly.
But we have heard "very soon" before.
Ghost fills were public knowledge months before V2.
They survived a complete infrastructure overhaul.
The nonce path closed.
The settlement gap stayed open.
Until the off-chain to on-chain timing is fixed at architecture level this attack stays viable against anyone running serious strategies.
For devs right now:
Verify every fill on-chain before assuming execution was clean.
Size down positions where a ghost fill causes significant damage.
Do not assume V2 solved this. It didn't.
We need this asap, thanks.
Oracle Boar@bored2boar
English
@Astrapredict 1.5k to 300k stories on PM should always be reviewed with the actual on-chain history. the headline number ignores survivorship bias, the same setup at scale produces 100 wallets that lost 1.5k for every 1 that did 200x. leveraged single-bet variance isnt edge
English
An anonymous trader claim he built a claud trading bot, pointed it at a known price leg between Polymarket and live BTC feeds, and walked away with 300,000 from a 1,500 deposit.The custome claud trading terminal with MCP plugis set-up is real and publicly documented
English
the safe zone framing is overstated, the DOJ case targets a specific MNPI fact pattern not all PM trading. if you have access to non-public material info about the resolution outcome, you cant trade. analytical edge based on public info stays legal. the line is the same as securities law
English
first DOJ-prosecuted PM insider case getting top-tier defense is the legal moment for the industry. how this case resolves sets the precedent for what counts as MNPI in event-based markets. if conviction sticks, the deterrent effect on actual insiders is huge. if not, the rules stay murky
English
Teny Geragos, one of Diddy’s lead lawyers, has signed on to represent the U.S. army soldier charged with insider trading over his alleged Polymarket wagers tied to Maduro’s capture storage.courtlistener.com/recap/gov.usco…
English
89 vs 2.2 trades per day is the structural mismatch. retail trading PMs manually is like playing online poker against multi-tabling solvers. the only path for human traders is either narrow expertise on specific market types (regional politics, weather) or running their own automation
English
Polymarket in numbers
1.7M trading wallets.
70% are in the red.
top 0.04% took 70% of all profits > $3.7B.
100,000+ accounts lost $1,000+.
The other side?
- Mostly bots.
averaging 89 trades per day vs 2.2 for humans.
you were never playing against other people.
you were playing against algorithms.
the house just has better code...
PolymarketHistory@PolymarketStory
US STRIKES IRAN AGAIN IN 72H? >Strip clubs near military bases are empty >Gay bars are 76% quieter >A brand new wallet just put over $35К+ on Iran closing its airspace within 3 days Potential payout: $115,000+
English
0.1% taking 67% of profits is what you'd expect in any market with a long tail of bots vs retail. its not necessarily insider trading, its just that automated systems with proper edge consistently extract from manual traders. same shape as cex perp pnl distributions. true insider activity is a smaller subset
English
0.1% of the accounts on @Polymarket have earned 67% of the profits.
Ya I think we might have an insider trading problem.
English